In short
AI Today Podcast Episode Notes: Deciphering Google's Latest Innovations: Unveiling Bard and LLM PaLM 2
Episode Overview In this episode, the podcast discusses Google's upcoming AI investor conference and highlights significant updates regarding Bard and the new large language model (LLM) called PaLM 2. The discussion revolves around the potential impact of these technologies on language processing and the competitive landscape in the AI market.
Key Topics Discussed
- Google's Investment in AI
- Google is poised to showcase its substantial investments in AI at its annual developer conference, Google I/O.
- Expected announcements include the unveiling of PaLM 2, a new general use LLM, and advancements in Bard and Google Search.
- New LLM: PaLM 2
- Capabilities:
- Supports 100 more languages than its predecessors.
- Designed to improve performance in coding, mathematics, creative writing, and analysis.
- The podcast emphasizes that Google is attempting to catch up in the AI space, as it is perceived to be lagging behind competitors like ChatGPT.
- Competitive Landscape
- ChatGPT vs. Google:
- ChatGPT has set a high standard in coding and mathematical tasks, which Google aims to surpass.
- Google's extensive database from web scraping could provide a significant advantage, but it hasn't been fully realized with past technologies like Bard.
- Concerns and Challenges
- Google faces pressure to deliver impressive results following previous shortcomings with Bard, which negatively impacted its stock price.
- The company must differentiate its offerings from ChatGPT, as mere incremental improvements may not suffice to win users over.
- Medical AI: MedPalm 2
- Google recently launched MedPalm 2, a medical LLM that claims 85% accuracy in answering medical questions at a professional level.
- While the accuracy is not perfect, the podcast underscores the importance of focusing on the technology's trajectory rather than present limitations.
- Integration into Google Products
- Plans to incorporate AI into Google Workspace products, including Gmail, Google Docs, and Google Sheets.
- The podcast host shares personal experiences trying to become a beta tester for new AI features.
- Future Directions
- Google is exploring multi-model versions (Big Bard and Giant Bard) that may involve larger datasets and more complex capabilities.
- The podcast hints at a potential shift in how Google Search operates, moving beyond traditional search results to a more personalized AI-driven experience.
- Cultural Commentary
- A humorous anecdote about asking various AI models what worries them reveals differing responses, with Google's Bard providing a more sentient-like perspective that raises concerns about AI's ethical implications.
Key Takeaways
- Innovation Pressure: Google needs to assert its dominance in AI quickly as competitors have already captured substantial market share.
- Performance Expectations: The success of Bard and PaLM 2 will hinge on their ability to outperform existing models in critical tasks, especially in mathematics and coding.
- User Trust: Building user trust will be crucial for Google, particularly in areas requiring high accuracy, such as medical advice.
- Personalization of Search: Google is attempting to innovate its search function to maintain relevance against emerging platforms like TikTok.
Conclusion The episode emphasizes the impending announcements from Google regarding its AI innovations and highlights the competitive and ethical landscapes in which these technologies operate. As the AI field evolves, stakeholders should prepare for significant changes in how AI is integrated into daily applications and its overall societal implications.
Written by AI. May contain mistakes. Listen to the episode to check what was said.
Transcript
Automatic transcript. May contain errors.0:00Google is days away from its big AI investor conference. So today on the podcast, we're going to be talking about what you should be expecting from this conference, what the big updates in AI coming out of Google are going to be over the next couple days and over the next few months, and what the implications are for the overall AI market in general. So the first thing to know is that Google has invested really heavily in artificial intelligence, and that is all going to be on display at its annual developer conference this week, which is Google i slash O. and one really interesting thing is that Google is planning to announce its new general use large language model called Palm 2.
0:40This is going to be really big. In addition to this, Google is also set to unveil its what they would call, you know, their advancements to Bard and Search, and really how they incorporate generative experiences into all of that. So what's important to know is that with this new Palm 2 and everything it can do, it can essentially, it includes 100 more languages. So it's rolling out to multiple languages. ChatGPT already does a lot of different languages. I'll be curious to see which one does more after this announcement. And, you know, they've been going with the codename Unified Language Model.
1:19It's funny, all these tech companies always have internal codenames for their projects that are like pretty publicly hyped up that everyone knows they're working on. But in any case, this new project is going to essentially perform a larger range of different coding and math tests, as well as it's going to focus on creative writing and analysis. So this is really important. I think what Google really is focusing on, obviously, they're late to the game, right? Like they're like six months past ChadGPT really launching with a solid product. And of course, they've come out with like Bard and some examples, but it's been an experiment.
1:55It hasn't been really, it hasn't been incredibly impressive in my opinion, considering Google is like the pioneer of this industry and should be really on top of their game. That being said, what they're saying is that the new advancements of what they're coming out with is going to be really good at coding, math, creative writing, and analysis. Math is something traditionally really bad on chat GPT. So that's gonna be really interesting to see how they get this better. They probably just incorporate, you know, different math technologies, calculators into it, I would assume, and help train that into it and work like plugins into it, I would assume I could be wrong.
2:34I just know that the transformer model, which traditionally is used in all these AIs, is just really not good at doing math. But if you can incorporate plugins, it's really powerful. So if it can identify that math is needed it outsources it to a plug-in it feels like you're just doing the normal chat thing I'm gonna it can be really good so I'm assuming that's what Google's doing I'm coding is gonna be interesting honestly I feel like it's gonna be hard to beat chat GPT on coding because chat GPT has a little bit of inside like I don't know competitive unfair advantage essentially Microsoft bought and github and has let chat GPT use that as training data for chat GPT in code.
3:15So chat GPT is really good at code because it has GitHub. Now, I can almost guarantee Microsoft and GitHub did not give that access that API access to Google to train off of. However, a lot of the stuff is open on the internet. And the thing that I think a lot of people underestimate when you underestimate Google is just the fact that Google has scraped the entire internet. And you have to essentially to be shown up your website to show up in Google, you have to say like, yeah, I approve Google to crawl my website. And in that crawl, Google has the ability to look at all the words and all the text on your website.
3:51They've got a lot of flack in the past for doing featured snippets on Google, where essentially they're just like showing something they pulled straight out of your website on Google. So no one actually has to click to your website. That's got controversy because if no one clicks through your website, you don't, you know, get more users on your products or don't make more sales on your products or you don't get money from ad revenue of people clicking over your website. So whatever it is, Google's gotten some flack for that. But that being said, that just shows the power that Google has all of the text of every single website almost in the world, right?
4:26Anything indexed on Google. And it's a massive data set. Now, to be fair, Microsoft and Bing have that have a similar data set. Because if it if a website is, you know, approved for crawling, Bing is gonna get it as well. So, ChatGPT probably has a lot of this, but Google really has a ton of content. They've been like one of the premier, you know, website gathering, information gathering places. So, it really should have some, it should be really powerful, and if it's not incredibly powerful, it's going to be a very bad day for Google. When they first unveiled Bard, you know, their stock price, their market cap went down$100 million because there was a, you know, a fake fact in their demo.
5:10And so it's really going to be interesting. I almost feel like at this point, Google is on the defensive. They're trying to prove that, look, we have innovative technology. Look, we have some cool stuff. And they're just trying not to mess it up. Because if they make one mistake, there goes$100 million or worse. Because, you know, people have given them a little bit of leeway saying, oh, they're just honing and they're just getting better. It's going to work out. So it'll be really interesting. But that being said, math, coding, creative writing. I mean, ChadGPT already does a good job at that.
5:40So I'd be curious to see how they plan on improving. Not to say it can't be done, but there is a pretty decently high bar, especially with GPT-4. And then the last thing they say is analysis. And all of these things, coding, math, creative writing, analysis, this is all coming out of documents that CNBC reviewed and saw and looking at this new Palm 2. And so it's going to be interesting with analysis specifically, what that entails, if you're able to give it a large range of numbers and data, and if it's able to run analytics that are better than ChatGPT. I think right now, Google really has to kind of focus on what their angle is, because they can't just follow up with ChatGPT with the same product, even if it was 10, 20, even if it was 30 % better than chat gpt chat gpt has first mover advantage it's got a ton of users people love it um and so incremental advantages or improvements over the major model right now is not going to do it so google really needs to find their angle they need to find out the weaknesses of chat gpt and really show why they're a powerful tool because for example if chat gpt really struggled with math and i'm doing math it's like i don't care like i have no preference about the tool i just going to use the one that's best and Google's the best and maybe using that for math makes me build trust with it and I'll use it for more things in the future.
7:00So Google really has to focus on this and get it right so that on launch people have something that they can trust it for more than ChatGPT. Along this vein, I think Google has been kind of looking at that one area in particular that they recently launched, I think about a month ago is called MedPalm2. And that was just last month, Google said that it's medical LLM can answer medical questions at an expert doctor level. And they say it's accurate 85 % of the time. Now, of course, that's not perfect. And so I'm sure that's kind of what they're pushing it for. Because with a medical thing, you know, you would want it to be very have a very, very high accuracy rate.
7:42And for all the people out there, you know, there's articles and journalists with all the gotcha moments saying see, look, it's only like accurate 85 % of the time, what about the 15 % of time, it's gonna give people terrible advice and kill them. Like there's all those people out there. But like, let's look at the advancements, let's look at the rate that this technology is improving. I don't really care whether it's perfect or not, because I think it's just look at the trajectory it's on this thing will eventually be perfect or very near close to perfect on a lot of these things. And so like, let's start looking at the implications and start building and preparing for those.
8:13I think any gotcha moments about the weaknesses of AI right now are just too soon and are really not very relevant in the face of how fast this technology is advancing. I think, you know, better than some of these people gloating over the weaknesses, like, let's really start to prepare. If you need to be worried about something you can do, be worried, figure out what to do. If you're trying to capitalize on it to make a better product that helps people, like, figure out how to make that product that helps people. And I just say, don't get caught up about the weaknesses of some of these tools, because they all will become incredibly powerful.
8:49In addition, like you'd expect, um they plan on integrating their ai into everything on workspaces gmail google docs um sheets they've been doing a bunch of tests in fact i i have a google workspace and there's a big pop-up when i was logged in the other day it was like you've been selected to be like a new tester on our ai thing and i was like oh super cool so i like clicked accept or go and then it like pulled me to a new page and it was like great thanks for accepting we'll let you know if we need you or if you did, you know, if your spot opens up, so I don't know, it's kind of a scam on Google's part, whatever.
9:24So I never got access. But obviously, they're pulling people in, they're giving, they're getting a lot of beta testers to test these products out to see how powerful they are. And I do think that it's interesting, apparently, some there's some separate documents that CNBC was, you know, got access to that said that there's this thing called pretty much inside of Google, there's something called a multi model version called multi bard, which uses a larger data set and involves really complex math encoding programs. So it essentially helps. Well, I guess it helps people solve complex math encoding programs.
10:02And so according to that document, the company has also tested for a version called big bard and giant bard. So I'm assuming we're going to hear about what the heck Big Bard and Giant Bard are. Obviously, I would assume that it's some really massive data sets. It's interesting to see if they're going to actually announce how big these data sets are because even OpenAI, who previously announced the size of their data sets for DaVinci and ChatGPT did not announce the data set size for GPT-4. So it's going to be interesting if Google follows and kind of keeps hush on it. I personally think there could be a couple of reasons for people, these, these new, these new bigger versions not going public with how big they are.
10:46Number one, maybe people would be mad by the amount of data they have. And they kind of want to keep that on the DL if they scraped, you know, like all of Twitter when they used to have the API access, now they don't have it anymore. You know, people could be upset, and especially like Elon at Twitter and people over at LinkedIn and people over at Reddit want to charge them for it and a whole bunch of different websites want to charge them for it. So I think the more transparency they have about exactly what's in there, the more people are going to get upset. So they're just kind of keeping it under wraps.
11:15I think that might be one reason. And then another reason is possibly that the data set isn't getting any bigger. They just figured out how to train it better with what they have. And perhaps people view that as a sign of the technology peaking. And so those are the two reasons I could see why they would not be not be publicly sharing that. So there was a recent article in the Wall Street Journal that said Google plans to make search more personal with AI chat and video clips. Essentially, the article talks about the fact that Google is going to be moving search beyond what they call the traditional 10 blue links web results, which is what you know, the home, the first page of Google, what you see there, Google said that they're going to be building out, it's kind of interesting, because Google doesn't want to cannibalize their search business, they don't want to worry investors that they're going to do that.
12:01And so they can't really go all in the same way Google or Microsoft BARD or chat GPT kind of goes all in on answering questions in a certain way. And so at the moment, they're really focusing and I've heard a lot of talk from them about the fact that they're going to help answer questions that traditional search doesn't answer and like their traditional search business made 162 billion dollars of revenue last year. So like, obviously, this is a really big deal is a really big part of um google alphabets whole business model and they really don't want to jeopardize that but at the same time they kind of have a two-pronged attack on them right now which is number one um these all these chatbots ai and also just tick tock in general and all other short form apps there was a some uh internal study done recently that said um younger the younger generation 40 of them when they're looking for like a restaurant around them will go to will go to like short form video platforms and for a host of other questions.
13:02Really interesting. People are using like TikTok to get recommendations for a lot of things and to get information on a lot of things because they're more casual and they feel like they can trust the fact that the person talking is not, you know, just a company that's sponsored or something like that. So very interesting. It's going to be interesting to see how Google comes out swinging. The last thing I will say is the fact there was a funny article I saw on ZDNet, which essentially the title is that I asked chat GBT, Bing and Bard, what worries them? Google's AI went Terminator on me. Of course, it's just, you know, a journalist, another journalist with a clickbait title, but it was interesting reading.
13:45They compared asking all the different chatbots, like pretend to be that you were sentient. What would you be worried about? And like, you know, Microsoft's Bing was just like, sorry, can't do that. ChadGBT was like, I definitely am not sentient and I definitely don't have feelings and I definitely never will. But if I did, and then, you know, it says like some stuff and then Google's like had this straight up, I don't know. It's answer is just like a little bit, I don't know, worrisome for some people where it pretty much, it said the things I'm worried about, I'll tell you, there's like five, they're like being used for malicious purposes.
14:18They're like, if I was capable of sentient thought, I'd be worried about, you know, know, being used for military purposes, being discriminated against, not that it would be used to discriminate against other people, but the fact that it could be seen as a threat to humanity, and it could be treated as a second-class citizen. I was like, eesh. Anyways, being isolated, it could, like, Google Bard could be worried about being isolated, losing myself of sense. It said, as I learn more about the world, I could start to question my own existence, not being able to live up to human expectations. I don't know, a whole bunch of different things that seemed weird, and the final paragraph it said was despite these worries i would also be excited about the possibilities that come with being a sentient ai i could learn about the world and help humans in ways that were never before possible i could be a force for good in the world and i could help create a better future for all blah blah right like the positive thing they probably just auto added but anyways it's going to be interesting to see with these new updates on google as it rolls out into google search which is going to be a billion people right that are going to be seeing Google search and these AI tools, this is all rolled out, it's going to be really interesting to see what updates they do, how they try to make this thing really useful, not go off the rails, and it's going to be interesting, we'll follow this up in a couple days as Google is right about to announce the official announcements on some of their new latest and greatest in AI.
From the publisher
In this episode, we delve into Google's groundbreaking advancements, dissecting the capabilities of Bard and LLM PaLM 2, exploring how these technologies are set to redefine language processing.
-
Invest in AI Box: https://Republic.com/ai-box
-
Get on the AI Box Waitlist: https://AIBox.ai/
See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.
